Speaker
Description
Forest aboveground biomass (AGB) is an important indicator of forest carbon storage and plays a key role in carbon monitoring and forest management. The increasing availability of spaceborne synthetic aperture radar (SAR) observations from Sentinel-1 and NISAR provides new opportunities for large-scale AGB estimation. In this study, a physics-based AGB retrieval framework is developed using both the Water Cloud Model (WCM) and a zero-baseline formulation of the Interferometric Water Cloud Model (IWCM). The WCM is used to model the backscattering response, while the IWCM exploits interferometric coherence to characterize the contributions from the forest canopy and underlying ground. By neglecting the vertical decorrelation terms associated with the perpendicular baseline, the IWCM is simplified and the number of unknown parameters is reduced. Combined with polarimetric optimization, three IWCM formulations are investigated to represent ground-only, forest-only, and combined ground–forest contributions to interferometric coherence. Model parameters are estimated without using reference AGB, which is introduced only for independent validation.
Two parameter estimation strategies are explored: nonlinear least-squares optimization and a deep-learning-based framework. In the latter, a neural network predicts the unknown physical model parameters from SAR observations, which are then passed through the WCM/IWCM forward models to reconstruct the observed backscatter or coherence. The network is trained by minimizing the discrepancy between modeled and observed SAR measurements, while physical constraints are imposed to ensure physically meaningful solutions. The proposed framework combines the interpretability of physical models with the flexible parameter estimation capability of deep learning for forest AGB retrieval.